Med Researcher Agent
lena-miyamoto/med-researcher-agent
Agent para pesquisa médica com arquivamento e análise de literatura científica
Ver no GitHubAgent reutilizável para pesquisa médica configurado para GitHub Copilot e Claude Code. Inclui fluxo local med-db para arquivar literatura estruturada de PubMed, Europe PMC e fontes de descoberta web, com skills para verificar afirmações médicas contra literatura científica e gerar relatórios de evidência estruturados.
Instalação
npx convez-ai lena-miyamoto/med-researcher-agent1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Verificar reivindicações médicas contra literatura científica
- Arquivar e catalogar literatura de PubMed e Europe PMC
- Gerar relatórios estruturados de evidência com vereditos
- Criar novas skills e agents compartilhados para pesquisa
Veredito

Solução robusta e bem arquitetada para pesquisadores médicos que precisam automatizar busca e análise de literatura. Requer familiaridade com Python, uv e conceitos de agents, mas oferece framework reutilizável de qualidade produção com sincronização cross-plataforma (Copilot/Claude).
README
A tradução para o português deste README ainda está na fila.
med-researcher-agent
Reusable medical research agent configuration for both GitHub Copilot and Claude Code.
The repository includes a local med-db/ workflow for archiving structured literature from PubMed, Europe PMC, and web
discovery sources (Google Scholar, DOAJ, Open Science Directory, Free Medical Journals, OpenMD, Trip Database).
Skill Surfaces
These are the main user-facing entry points in this repository.
| Skill | What it does |
|---|---|
analyze-med-claims | Verify medical claims against the scientific literature and produce structured evidence reports with verdicts. |
create-med-skill | Add a new shared skill with .agents as the source of truth plus .github and .claude wrappers. |
create-med-agent | Add a new shared agent with .agents as the source of truth plus .github and .claude wrappers. |
optimize-repo | Audit and clean up repo customization files for consistency and source-of-truth hygiene. |
All skills are defined under .agents/skills/ and surfaced through thin .github/skills/ and .claude/skills/
wrappers.
Setup and Tests
The project uses uv for Python dependency management and repo-standard command execution.
uv is required for running skills and repo tooling in this repository.
Installing uv
# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows (WinGet)
winget install --id=astral-sh.uv -e
# Cross-platform via pip / pipx
pipx install uv
After installing uv, sync the local environment before running tests or repo tools:
uv sync
Running tests
# Run the full suite via the project script
uv run test
# Or pass through specific pytest arguments
uv run test tests/test_med_db_validate.py -q
Running med-db tools with uv
uv run med-db --help
uv run med-db --query 'endometriosis' --archive-first 5 --validate
uv run med-db-validate --med-db med-db
uv run med-db-lookup --pmid 12345678
uv run med-db-query --list-topics
These are the canonical repo commands. Use them instead of direct python3 .agents/scripts/... execution.
Adding or extending Python scripts
# Add runtime dependencies to pyproject.toml, then:
uv sync # sync lockfile and venv
uv run test # re-run the full suite
Instruction Layout
AGENTS.md: repo-wide, tool-agnostic workflow.CLAUDE.md: Claude-specific routing..github/copilot-instructions.md: Copilot-specific routing..agents/agents/med-researcher.md: shared med-researcher role and output rules..github/agents/med-researcher.agent.md,.claude/agents/med-researcher.md: thin med-researcher harness wrappers..agents/skills/<name>/SKILL.md: shared cross-harness skill procedures..github/skills/,.claude/skills/: thin harness wrappers pointing at the shared skill..agents/scripts/: local archive tooling used by both agents.
Agent wrappers stay aligned; differences are limited to harness frontmatter, tool names, and model selection.
Claude Code Usage
Start a session with the agent directly:
claude --agent med-researcher